Special flight control method for hydrogen energy unmanned aerial vehicle

Abstract

The invention relates to the technical field of path control, and discloses a flight control method special for a hydrogen energy unmanned aerial vehicle, which comprises the following steps: carrying out dimensionless time alignment on a multi-source sensing state, and extracting an inverse gravity drainage exposure memory amount reflecting historical accumulation of a bad attitude; the energy supply fragility degree of a future path section is predicted by utilizing inverse gravity drainage exposure memory amount feedforward, the power proportion of a battery and a super capacitor is allocated in advance according to the energy supply fragility degree, and meanwhile, the hydrogen emission frequency is regulated and controlled. In addition, the memory amount and the energy supply gap are jointly incorporated into the path optimization cost to remodel the safe flight route and travel speed. According to the method, time dislocation between transient obstacle avoidance and bottom layer energy response is eliminated, and the long-endurance operation robustness of equipment under complex working conditions is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of path control technology, and more specifically, to a flight control method specifically for hydrogen-powered unmanned aerial vehicles (UAVs). Background Technology

[0002] Hydrogen-powered drones equipped with proton exchange membrane fuel cells and supercapacitors have significant advantages in the field of long-endurance monitoring. However, current flight control and energy management are usually independent of each other. Flight control mainly generates thrust commands based on real-time spatial attitude deviations, while battery management passively follows instantaneous power loads to provide power.

[0003] During missions involving sustained attitude changes such as steep climbs, continuous hovering, and obstacle avoidance, the internal directional flow channels of the battery may remain in an orientation unfavorable to water drainage for extended periods. Furthermore, due to the inherent lag in the system's exhaust and oxygen supply compressors, this obstruction of gas and liquid drainage caused by the spatial attitude will continuously accumulate within the battery stack. Even after the drone returns to level flight, the adverse conditions such as oxygen starvation and localized water stagnation cannot be immediately eliminated.

[0004] Existing conventional algorithms do not address the adverse drainage effects accumulated from attitude history. When the system encounters sudden high loads or strong gusts while this latent water accumulation is not completely resolved, it is highly susceptible to a sudden drop in available energy supply that cannot match the external space flight control requirements. This severe disconnect between space maneuvering and internal energy response leads to loss of heading control, thereby weakening the system's safety and hazard avoidance capabilities during near-obstacle operations, posing a critical control challenge that urgently needs to be overcome. Summary of the Invention

[0005] This invention provides a flight control method specifically for hydrogen-powered drones, solving the technical problems mentioned in the background art.

[0006] This invention provides a flight control method specifically for hydrogen-powered drones, applicable to hydrogen-powered drones comprising a proton exchange membrane fuel cell, a supercapacitor, an oxygen supply component, a hydrogen supply component, a hydrogen exhaust component, a sensor unit, and a multi-rotor component, including: The sensor state quantities of the sensor unit are acquired and synchronized, and dimensionless processing is performed according to the feasible value range of the rolling window to obtain a dimensionless state vector. The path power demand index is then calculated by feedforward. Based on the aircraft attitude of the hydrogen-powered UAV and the favorable drainage direction of the proton exchange membrane fuel cell, the anti-gravity drainage exposure memory quantity, which reflects the accumulation of unfavorable drainage, is extracted. The power vulnerability of future path segments is predicted by using the memory amount of the anti-gravity drainage exposure and the oxygen sufficiency in the sensed state quantity, and then the future energy supply gap index is output. Based on the future energy supply gap index and the memory capacity exposed by the anti-gravity drainage, the target output of the supercapacitor and the target output of the fuel cell are generated, and the hydrogen supply regulation command and the continuous hydrogen discharge duty cycle are generated simultaneously and adaptively. The exposure memory of the anti-gravity drainage and the future energy supply gap index are incorporated into the path evaluation cost, and the reconstructed local path and the speed distribution along the path are generated by optimizing the candidate path set. The center of gravity position vector is updated based on the mass migration caused by fluid material consumption, and a feedforward compensation torque is generated based on the center of gravity position vector to correct the rotor target command. Assess the overall risk of the system and generate an emergency takeover coefficient. Use the emergency takeover coefficient to smoothly merge the reconstructed local path and the emergency landing path to generate the final command path.

[0007] The beneficial effects of this invention include: by extracting the adverse liquid discharge effect of the fuel cell caused by space maneuvering into an anti-gravity liquid discharge exposure memory, and integrating it throughout the power coordinated scheduling and route optimization process. By accurately predicting the contraction trend of this memory on subsequent available energy, the system can not only guide the aircraft to avoid flight segments that exacerbate severe liquid discharge, but also preemptively call upon supercapacitors to buffer transient load impacts. This closed-loop transformation from single passive following to bidirectional active control eliminates the potential for power cliff caused by historical attitude accumulation and gas supply lag. Combined with dynamic center of gravity offset compensation and continuous smooth takeover mechanism, this invention achieves consistency assurance between the energy metabolism system and complex three-dimensional space risk avoidance, significantly enhancing the long-term operational safety and anti-interference resilience of the hybrid power platform in harsh environments. Attached Figure Description

[0008] Figure 1 This is a flowchart of a flight control method for hydrogen-powered drones according to the present invention. Detailed Implementation

[0009] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0010] like Figure 1 As shown, a flight control method specifically for hydrogen-powered drones is applied to hydrogen-powered drones comprising a proton exchange membrane fuel cell, a supercapacitor, an oxygen supply component, a hydrogen supply component, a hydrogen exhaust component, a sensor unit, and a multi-rotor component, including: The sensor state quantities of the sensor unit are acquired and synchronized, and dimensionless processing is performed according to the feasible value range of the rolling window to obtain a dimensionless state vector. The path power demand index is then calculated by feedforward. Based on the aircraft attitude of the hydrogen-powered UAV and the favorable drainage direction of the proton exchange membrane fuel cell, the anti-gravity drainage exposure memory quantity, which reflects the accumulation of unfavorable drainage, is extracted. The power vulnerability of future path segments is predicted by using the memory amount of the anti-gravity drainage exposure and the oxygen sufficiency in the sensed state quantity, and then the future energy supply gap index is output. Based on the future energy supply gap index and the memory capacity exposed by the anti-gravity drainage, the target output of the supercapacitor and the target output of the fuel cell are generated, and the hydrogen supply regulation command and the continuous hydrogen discharge duty cycle are generated simultaneously and adaptively. The exposure memory of the anti-gravity drainage and the future energy supply gap index are incorporated into the path evaluation cost, and the reconstructed local path and the speed distribution along the path are generated by optimizing the candidate path set. The center of gravity position vector is updated based on the mass migration caused by fluid material consumption, and a feedforward compensation torque is generated based on the center of gravity position vector to correct the rotor target command. Assess the overall risk of the system and generate an emergency takeover coefficient. Use the emergency takeover coefficient to smoothly merge the reconstructed local path and the emergency landing path to generate the final command path.

[0011] Preferably, the sensing state quantities of the sensor units are acquired and synchronized, and dimensionless processing is performed according to the feasible value range of the rolling window to obtain a dimensionless state vector. The path power demand index is then calculated using a feedforward method, including: Current perceptible distance Divide by current flight speed With control cycle The product plus the numerical stability term The sum, rounded up to determine the preview distance from the walk number To generate the scrolling window: Using time interpolation combined with the numerical stability term The asynchronously acquired sensing state quantity Align to the current control moment: The aligned perceived state variables are normalized according to the feasible value range of the scrolling window to form the dimensionless state vector. The dimensionless state vector Includes normalized speed Normalized tangential acceleration Normalized vertical acceleration Normalized trajectory tracking error Normalized path curvature Normalized roll angle Normalized pitch angle Normalized hydrogen pressure Normalization pressure risk Normalized average stack temperature Normalized temperature dispersion Normalized temperature risk Normalized stack current Normalized oxygen supply adequacy Normalized supercapacitors can be used for state of charge. Normalized local hydrogen concentration risk Normalized barrier distance and normalized barycenter position vector : The normalized speed With the normalized tangential acceleration The product of, the normalized speed The square of the normalized path curvature The product of the normalized vertical acceleration and the normalized trajectory tracking error The sum of the two demands is used to obtain the demand numerator. The sum of the two demands is then used as the demand denominator. Finally, the demand numerator is divided by the demand denominator to obtain the path power demand index. : Sensing state variables are a multi-source set of fundamental states used to characterize the flight state, energy state, and environmental state of an unmanned aerial vehicle (UAV). These variables can be obtained through a combination of inertial measurement units, satellite positioning modules, barometers, hydrogen pressure sensors, fuel cell temperature sensors, current sensors, oxygen supply monitoring units, supercapacitor management units, hydrogen concentration sensors, and lidar or millimeter-wave radar.

[0012] The current perceptible distance is the effective perceptible distance that can be stably identified and used for planning along the flight path at the current control moment. It can be obtained by measuring the nearest effective obstacle boundary or the farthest effective passable boundary within the flight path sector using forward-looking lidar, millimeter-wave radar, or depth vision.

[0013] Current flight speed is the speed of the UAV relative to the ground or local navigation coordinate system at the current moment. It can be obtained through a combination of speed output from the satellite positioning module, visual odometry, or airspeed measurement components.

[0014] The control cycle is a fixed time interval between two consecutive control updates. A value of 0.05 seconds is preferred, as this balances flight control bandwidth, real-time energy management, and multi-source sensor refresh rate, ensuring stable system operation.

[0015] The numerical stability term is a small positive value used to prevent the algorithm from diverging under conditions such as a denominator of 0, excessively small time differences, or extreme conditions. The preferred value range is 0.000001 to 0.001, which can suppress numerical singularities without significantly disturbing the main calculation results.

[0016] The continuous preview value is the length of the continuous preview obtained by adjusting the current perceived distance according to the current flight speed and control cycle.

[0017] The preview step count is the number of future prediction steps obtained by rounding up the continuous preview value.

[0018] A rolling window is a finite prediction interval that expands forward along a local prediction path starting from the current control moment.

[0019] The current control time is the system time reference corresponding to the closed-loop calculation of flight control and energy management in this operation.

[0020] The previous sampling time is the valid sampling time point that is closest to and precedes the current control time.

[0021] The next sampling time is the effective sampling time point that is closest to and follows the current control time.

[0022] The aligned sensing state quantity is the set of synchronized states obtained by uniformly interpolating each asynchronous sampled state according to the current control time.

[0023] The feasible range is the upper and lower bounds of each state variable that are allowed to participate in dimensionless mapping under the current aircraft type, current mission, and safety packet. Preferably, it is a closed range that is extended outward by 10% to 20% from the normal operating range obtained by flight calibration. This value can cover normal disturbances and avoid abnormal spikes from compressing the normalization resolution.

[0024] A dimensionless state vector is a set of states formed by mapping different physical quantities to a uniform scale.

[0025] Normalized velocity is a velocity quantity that maps the current flight speed to a uniform dimensionless interval according to the feasible range of velocity values. It is used to reflect the degree of influence of flight speed on energy demand.

[0026] Normalized tangential acceleration is a dimensionless quantity obtained by mapping the acceleration along the tangential direction of the current path through the feasible value range. It is used to characterize the change in propulsion power caused by acceleration and deceleration.

[0027] Normalized vertical acceleration is a dimensionless quantity that is mapped vertical acceleration according to the feasible range. It is used to characterize the additional load caused by climb, descent, and aerodynamic disturbance.

[0028] Normalized trajectory tracking error is a dimensionless quantity that maps the deviation between the current position and the reference trajectory. It is used to characterize the additional control strength required to correct the trajectory.

[0029] Normalized path curvature is a dimensionless quantity mapped from the degree of local path curvature, used to reflect the impact of turning maneuvers on lateral loads and power requirements.

[0030] The normalized roll angle is a dimensionless quantity that maps the roll attitude of a UAV to the feasible range of attitudes, and is used to characterize the degree of lateral tilt of the airframe.

[0031] Normalized pitch angle is a dimensionless quantity that maps the pitch attitude of a UAV to the feasible range of attitudes, and is used to characterize the degree of longitudinal pitching or tilting of the aircraft.

[0032] Normalized hydrogen pressure is a dimensionless quantity that maps the pressure of the hydrogen storage or supply circuit according to the operating range, and is used to reflect the fuel supply margin.

[0033] Normalized pressure risk is a dimensionless risk quantity derived from the degree to which hydrogen pressure deviates from the safe operating range, and is used to characterize the abnormal risk on the hydrogen supply side.

[0034] The normalized average temperature of the fuel cell stack is a dimensionless quantity that is mapped from the average temperature of multiple temperature sampling points of the fuel cell stack. It is used to characterize the overall thermal state of the fuel cell stack.

[0035] Normalized temperature dispersion is a dimensionless quantity that maps the temperature differences between various temperature sampling points of the fuel cell stack, and is used to characterize the degree of local thermal non-uniformity.

[0036] Normalized temperature risk is a thermal anomaly risk mapping quantity that is jointly reflected by average temperature and temperature dispersion, and is used to characterize the risk of thermal imbalance of fuel cell stack.

[0037] Normalized stack current is a dimensionless quantity that maps the stack output current to the rated operating range and is used to characterize the current electrochemical load intensity.

[0038] Normalized oxygen sufficiency is a mapping of oxygen supply capacity relative to the current oxygen demand level of the fuel cell stack, used to characterize whether the air supply is sufficient.

[0039] The normalized state of charge (SOC) of a supercapacitor is a dimensionless quantity mapped to the current state of charge of the supercapacitor that can safely release energy. It is used to characterize the short-term power buffer margin.

[0040] Normalized local hydrogen concentration risk is a dimensionless risk quantity derived from the monitoring value of local hydrogen concentration in the body, used to characterize the risk of hydrogen leakage or local accumulation.

[0041] Normalized obstacle distance is a dimensionless quantity that maps the effective distance between the UAV and the nearest obstacle, and is used to characterize the proximity of obstacles and the urgency of obstacle avoidance.

[0042] The normalized center of gravity position vector is the result of scaling the position vector of the UAV's current integrated center of gravity relative to the body reference point, and is used to characterize the impact of mass transfer on attitude control.

[0043] The demand numerator is a comprehensive demand quantity composed of normalized velocity, normalized tangential acceleration, normalized path curvature, normalized vertical acceleration, and normalized trajectory tracking error. It is used to characterize the direct power demand intensity of the current path segment.

[0044] The demand denominator is a normalized denominator formed by adding 1 to the demand numerator, used to compress the total demand into a stable dimensionless range.

[0045] The path power demand index is a dimensionless index that represents the combined power demand for propulsion and attitude control in the current or preview path segment.

[0046] In practice, to generate the current perceptible distance, preview continuous values, preview outlier counts, and scrolling window, a forward perception corridor aligned with the current heading needs to be established in front of the aircraft's nose, prioritizing a fan-shaped field of view of 30 degrees left and right and 15 degrees up and down. After denoising, outlier removal, and confidence filtering of the forward-looking lidar, millimeter-wave radar, and depth vision results, the nearest effective obstacle boundary or the farthest effective passable boundary is taken as the current perceptible distance along the centerline of the local reference path. When the environment is open and the effective range of the sensors exceeds the task limit, the current perceptible distance is... The perceived distance is taken as the upper limit of the task; when there is severe near-field occlusion or too many low-confidence points, the current perceived distance is taken as the most recent stable value and maintained for 1 control cycle; the preview continuous value is obtained by dividing the current perceived distance by the product of the current flight speed and the control cycle and the sum of the numerical stability terms; the preview deviation number is obtained by rounding up the preview continuous value, and a lower limit of 10 steps and an upper limit of 120 steps are set. The lower limit is used to avoid insufficient look-ahead due to the window being too short, and the upper limit is used to limit the amount of real-time computation; the scrolling window expands point by point along the prediction path from the current control time and moves forward once in each control cycle.

[0047] In specific implementation, for the buffer structure and interpolation boundary settings for time alignment of asynchronous sensing state variables, a timestamped double buffer or ring buffer needs to be established for each sensing state variable, and at least the two most recent valid sample values ​​should be retained. When the current control time is between the previous and next sampling time, linear interpolation is used to complete the time alignment. When the interval between two sampling times exceeds three control cycles, it is determined that there is hysteresis in the data, and the previous valid value is retained first while the uncertainty of the state is increased simultaneously. When the next sampling time does not exist, the previous valid value is allowed to be extrapolated for a maximum of one control cycle. If this time limit is exceeded, degradation processing is triggered. When the sensing state variable belongs to discrete switch state, linear interpolation is not used, but the most recent valid value retention strategy is adopted.

[0048] In specific implementation, the determination of feasible value ranges and the setting of normalization rules for each state variable require first determining the physical lower limit, physical upper limit, alarm lower limit, and alarm upper limit of each state variable through ground calibration, bench testing, and actual flight envelope testing. For state variables that only take non-negative values, linear normalization is adopted with the lower bound mapped to 0 and the upper bound mapped to 1. For state variables with positive and negative directional significance, symmetrical interval normalization is adopted, so that the 0 value falls in the middle position, the negative limit corresponds to the lower boundary, and the positive limit corresponds to the upper boundary. For risk-type state variables, monotonic mapping is adopted with the safe side being 0 and the dangerous side being 1. Data exceeding the feasible value range is uniformly saturated and truncated to prevent abnormal peaks from damaging the overall scale. Speed ​​is normalized first from 0 meters per second to the maximum safe flight speed, roll angle and pitch angle are normalized first from -45 degrees to +45 degrees, trajectory tracking error is normalized first from 0 meters to the upper limit of allowable trajectory error, and path curvature is normalized first from 0 to the reciprocal of the minimum flyable turning radius.

[0049] In specific implementation, the determination of normalized pressure risk, normalized temperature risk, normalized oxygen supply adequacy, normalized local hydrogen concentration risk, and normalized center of gravity position vector requires the following steps: Normalized pressure risk is calculated from the deviation of the measured hydrogen supply pressure from the rated operating pressure window; the closer the pressure is to the lower or upper limit, the closer the risk is to 1. Normalized temperature risk is calculated as a weighted sum of the stack's average temperature and temperature dispersion; excessively high average temperature and large temperature differences both increase the risk. Normalized oxygen supply adequacy is determined by the ratio of real-time oxygen supply mass flow rate to the theoretical oxygen demand corresponding to the current stack current, and the portion exceeding the adequacy threshold is saturated. Normalized local hydrogen concentration risk is obtained by mapping the local hydrogen concentration in the cabin or body to the warning and alarm concentration ranges; a lower value is used when the concentration is below the warning threshold, and a higher value is used directly when the concentration is above the alarm threshold. The normalized center of gravity position vector is obtained by weighted summation of the local center of gravity positions based on the hydrogen storage mass, liquid water equivalent mass, supercapacitor mass, load mass, and body reference mass.

[0050] In practical implementation, for limiting, filtering, and handling outliers in the path power demand index, outliers need to be removed from the normalized tangential acceleration, normalized vertical acceleration, and normalized trajectory tracking error before they are included in the comprehensive demand calculation. When the comprehensive demand is less than 0, it is truncated to 0 to avoid misjudgment of negative power demand during deceleration or descent. When the comprehensive demand is too large, it is preferentially limited to the calculation range of 0 to 10 to maintain the stability of the subsequent normalized denominator. If any key state variable involved in the calculation fails, the path power demand index reverts to the previous valid value and a safety margin is added. For cases where the jump variable exceeds 0.2 within three consecutive control cycles, a first-order smoothing filter can be added to avoid power allocation command jitter.

[0051] Preferably, based on the aircraft attitude of the hydrogen-powered UAV and the favorable drainage direction of the proton exchange membrane fuel cell, the anti-gravity drainage exposure memory reflecting the accumulation of unfavorable drainage is extracted, including: Subtract the unit vector of the direction of gravity from one. The direction cosine matrix from the machine system to the fuel cell flow channel coordinate system and the unit vector of the favorable drainage direction The product of these values, divided by two, yields the normalized anti-gravity drainage exposure. : The negative control cycle The exponent value is obtained by taking the natural exponential function as the independent variable, thus yielding the memory decay coefficient. : Normalized time since the last hydrogen emission Divide by normalized recovery time and the numerical stability term The sum of these yields the normalized hydrogen emission process. : Exposing the memory capacity of the previous cycle's anti-gravity drainage. Multiplied by the memory decay coefficient Obtain the first product term; subtract the memory decay coefficient from one. The difference, the normalized antigravity drainage exposure The normalized stack current The normalized hydrogen emission process And one minus the normalized oxygen sufficiency The differences are multiplied together to obtain the second product term. The first product term is added to the second product term to obtain the anti-gravity drainage exposure memory amount for the current period. : The attitude of the drone system is the spatial orientation of the drone relative to the navigation reference frame.

[0052] The optimal drainage direction for a proton exchange membrane fuel cell is the direction in which the liquid water inside the stack is most easily discharged under the combined effects of gravity and airflow. Preferably, the installation direction is aligned with the main flow outlet and drain port of the stack, as this minimizes liquid water retention and reduces oxygen supply blockage.

[0053] The gravity direction unit vector is a standardized vector used to characterize the direction of gravity in the reference coordinate system.

[0054] The direction cosine matrix from the machine system to the fuel cell stack flow channel coordinate system is a fixed calibration relationship that transforms the attitude relationship in the machine system to the fuel cell stack flow channel installation coordinate system. It can be obtained through whole-machine assembly calibration, three-dimensional attitude measurement, and installation reference calibration.

[0055] Normalized antigravity drainage exposure is a dimensionless quantity representing how inconsistent the direction of the fuel cell flow channel is with gravity drainage at the current attitude, and is used to characterize the degree of adverse drainage at a single moment.

[0056] The memory decay coefficient is a coefficient used to describe the degree to which the impact of historical adverse drainage gradually decays over time. It is used to balance historical exposure with current exposure in the memory model.

[0057] The normalized time since the last hydrogen purging is the normalized result of the time elapsed from the end of the most recent hydrogen purging event to the current moment, and is used to characterize the extent to which the liquid water removal effect has decayed.

[0058] Normalized recovery time is a normalized parameter of the nominal time required for the gas-liquid distribution inside the fuel cell stack to return to stability after hydrogen discharge. It is preferably a recovery length equivalent to 1 to 5 seconds after conversion, which can cover the recovery process in scenarios with mild to moderate water accumulation.

[0059] The normalized hydrogen emission process is a dimensionless process quantity obtained by comparing the normalized time since the last hydrogen emission with the recovery time. It is used to characterize the stage to which the hydrogen emission effect has progressed.

[0060] The anti-gravity drainage exposure memory of the previous cycle is the accumulated adverse drainage history memory result from the previous control cycle, which is used to perpetuate the historical effect in the current cycle.

[0061] The first product term is the product of the anti-gravity drainage exposure memory and the memory decay coefficient from the previous cycle, used to characterize the portion of historical adverse drainage effects retained in this cycle.

[0062] The second product term is the result of the product of current adverse drainage exposure, stack load, hydrogen discharge process, and oxygen insufficiency, and is used to characterize the newly added adverse drainage accumulation in this cycle.

[0063] The anti-gravity drainage exposure memory is a cumulative memory obtained by integrating the effects of historical adverse drainage and the current newly added adverse drainage exposure. It is used to characterize the degree of historical residual obstruction of liquid drainage inside the fuel cell stack.

[0064] In practical implementation, regarding the establishment of the favorable drainage direction and the fuel cell stack flow channel coordinate system, it is necessary to first establish the mechanical system based on the geometric center of the UAV body, and then establish the fuel cell stack flow channel coordinate system based on the main flow channel direction, the water collection area direction, and the drainage outlet direction. The installation attitude of the fuel cell stack flow channel coordinate system relative to the mechanical system is determined through a three-dimensional assembly model, and then a water injection and drainage test is conducted on a ground test bench for verification. The direction in which liquid water is discharged fastest and leaves the least residue under the combined action of gravity and gas supply is defined as the favorable drainage direction. When the fuel cell stack adopts a parallel flow channel structure, the favorable drainage direction is preferentially taken as the direction from the flow channel outlet to the drainage outlet. When the fuel cell stack adopts a serpentine flow channel structure, the equivalent main direction with the highest comprehensive drainage efficiency can be preferentially taken.

[0065] In practical implementation, the coordinates and meaning of the normalized anti-gravity drainage exposure should be uniformly represented by a right-handed coordinate system, with the forward, rightward, and downward axes of the machine system as fixed axes. The gravity direction in the navigation reference system always points towards the Earth's center. The coordinates are then transformed into the machine system or fuel cell flow channel coordinate system through attitude calculation. When the gravity projection is consistent with the favorable drainage direction, the normalized anti-gravity drainage exposure should be close to 0. When the gravity projection is completely opposite to the favorable drainage direction, the normalized anti-gravity drainage exposure should be close to 1. When the two are orthogonal, the normalized anti-gravity drainage exposure should be at an intermediate level.

[0066] In practice, the setting of the memory decay coefficient and normalized recovery time requires conducting short-duration high-current impact, continuous climbing attitude maintenance, and periodic hydrogen evacuation tests on a ground test bench to record the time required for the stack voltage to recover, temperature difference to fall back, and oxygen supply to recover after hydrogen evacuation. Then, the time scale of memory decay is determined based on the recovery curve, so that the impact of historical adverse drainage can gradually decay over several control cycles rather than disappearing instantaneously. The normalized recovery time is preferably set to a recovery length equivalent to 1 to 5 seconds, and the memory decay coefficient should be updated in a consistent manner with the control cycle. When the UAV has a large load, a long flow channel, or a large drainage outlet resistance, the normalized recovery time can be appropriately increased.

[0067] In practice, regarding the updates to the normalized time since the last hydrogen venting and the normalized hydrogen venting process, the timer for the period since the last hydrogen venting needs to be reset to zero when the hydrogen venting valve is opened, and the accumulation should restart after the hydrogen venting is completed. The accumulated time should continuously increase in steps according to the control cycle and be normalized according to a preset time scale. The normalized hydrogen venting process is determined by the ratio of the accumulated time to the recovery time, and the value range is limited to 0 to 1. When the interval between two consecutive hydrogen ventings is too short, the residual recovery effect of the previous hydrogen venting should be retained to avoid model distortion due to frequent resetting. When hydrogen venting fails or the valve does not open as commanded, the timer should not be reset, but should be marked as an abnormal hydrogen venting event.

[0068] In specific implementation, for the initialization, limiting, and reset of the anti-gravity drainage exposure memory, the anti-gravity drainage exposure memory should be set to 0 or the measured low value under stable level flight conditions before takeoff during system power-on initialization; after each control cycle update, it should be limited to between 0 and 1 to prevent cumulative overflow caused by sensor malfunctions; when the UAV continuously maintains a level flight attitude conducive to drainage, with sufficient oxygen supply, temperature difference reduction, and at least one effective hydrogen evacuation, the memory value can be gradually reduced; when a temperature sensor failure, oxygen supply sensor failure, or abnormal hydrogen evacuation execution is detected, the current memory value should be frozen and the safety margin increased; when the system performs a landing shutdown or the fuel cell stack completely stops, the memory value should be reinitialized on the next startup.

[0069] Preferably, the power vulnerability of future path segments is predicted by utilizing the anti-gravity drainage exposure memory and the oxygen sufficiency in the sensed state quantities, thereby outputting a future energy supply gap index, including: Using local velocity estimates Temperature at each sensing point Perform sensitivity weighting The normalized temperature dispersion is obtained by correcting the difference between the corrected maximum and minimum values. : Subtract the memory amount of the anti-gravity drainage exposure from 1. The difference, minus the normalized temperature dispersion The difference, the normalized hydrogen pressure and the normalized oxygen supply adequacy By multiplying the products consecutively, taking the fourth root of each product, and then subtracting the fourth root from the product, we obtain the normalized path segment power vulnerability. As for the aforementioned power vulnerability: Subtract the normalized path segment power vulnerability from 1 The difference multiplied by the current normalized fuel cell output The available power of the fuel cell in the predicted time domain is obtained. : For each discrete point of the preview path segment Calculate the corresponding path power demand index as the predicted load demand. The predicted load demand Divide by the available power of the fuel cell Supercapacitors can support power and the numerical stability term The summation of these values ​​yields the future energy supply gap indicator. : The local velocity estimate is the result of estimating the speed of airflow at different locations on the fuel cell stack, and is used to correct the sensitivity of each temperature sampling point to thermal inhomogeneity.

[0070] The temperature at each sensing point is a set of temperature samples placed at different locations on the fuel cell stack, used to characterize the thermal distribution of the stack. This temperature can be acquired through distributed thermocouples, thermistors, or integrated temperature sensors.

[0071] Sensitivity weight is a correction weight used to characterize the contribution of different sensing points to thermal risk under local flow velocity changes. The preferred value range is 0.1 to 0.6, which reflects the actual characteristics that low flow velocity areas are more likely to form hotspots and oxygen-depleted areas.

[0072] The corrected sensor point temperature is a temperature value obtained by correcting the original sensor point temperature together with the local flow velocity sensitivity weight, and is used to more realistically reflect potential hotspot risks.

[0073] Normalized path segment power vulnerability is a dimensionless vulnerability obtained by integrating adverse drainage memory, temperature dispersion, hydrogen pressure, and oxygen supply adequacy. It is used to characterize the risk of future shrinkage of available power in fuel cells along the path segment.

[0074] The current normalized fuel cell output is a dimensionless result of the actual output power of the fuel cell at the current moment mapped to its rated capacity, and is used to predict the available power of the fuel cell in the short term.

[0075] The prediction time domain is the time interval extended from the current control moment to the future for assessing the available power and energy supply gap of the fuel cell. It is preferably 0.5 seconds to 2 seconds, which can cover local obstacle avoidance and short-term maneuvers while maintaining the real-time nature and controllability of the prediction results.

[0076] The available power of a fuel cell is the power capability that the fuel cell can still stably provide in the predicted time domain after taking into account power vulnerability.

[0077] The preview path segment is a future local path segment within the scrolling window that is to be evaluated.

[0078] Predicted load demand is a path power demand indicator that previews the discrete points of a path segment, used to represent the load level required at different path locations in the future.

[0079] The power that a supercapacitor can support is the short-term power that the supercapacitor can release, estimated based on the current state of charge, voltage, temperature, and allowable discharge capacity.

[0080] The future energy supply gap indicator is the gap obtained by comparing the predicted load demand with the available power of fuel cells and the power that supercapacitors can support.

[0081] In practical implementation, to establish the estimated local flow velocity and sensitivity weights, discrete flow nodes need to be established at the inlet, outlet, central area, and edge area of ​​the fuel cell stack. The local flow velocity of each node is estimated by comprehensively considering the oxygen compressor speed, air flow rate, flow channel pressure drop, fuel cell stack current, and structural location. Areas with lower flow velocities are more prone to heat accumulation and oxygen depletion, so they should be assigned higher sensitivity weights. Sensitivity weights can be calibrated through bench thermal imaging tests and multi-point temperature comparison tests, and segmented values ​​are used in different load ranges. When the local flow velocity is lower than 60% of the rated average flow velocity, the correction amount should be significantly increased; when the local flow velocity is close to the rated average flow velocity, the correction amount can be reduced to a smaller level.

[0082] In practical implementation, for the temperature layout of each sensing point and the calculation of normalized temperature dispersion, at least 6 temperature sampling points need to be set up on the fuel cell stack, preferably near the anode inlet, near the cathode inlet, in the middle of the flow channel, near the flow channel outlet, at the edge of the stack, and in potential hot spots. The sampled values ​​are first synchronized with time and outlier removal, and then combined with the local flow velocity estimate for sensitivity weight correction. The normalized temperature dispersion is defined as the mapping result of the difference between the highest and lowest temperatures after correction relative to the upper limit of the allowable temperature difference. When the temperature difference is lower than the normal average temperature threshold, the normalized temperature dispersion takes a low value; when the temperature difference is close to or exceeds the upper limit of the allowable temperature dispersion, the normalized temperature dispersion takes a high value.

[0083] In practical implementation, for the mapping and limiting rules of normalized path segment power vulnerability, this quantity needs to be used to represent the probability and degree of potential contraction of fuel cell available power in future path segments, with the value range limited to between 0 and 1; its constituent factors should simultaneously reflect adverse drainage history, thermal non-uniformity, hydrogen pressure margin, and oxygen supply sufficiency; the fourth-order processing is used to weaken the abrupt impact of extreme changes in a single factor on the overall result, so that vulnerability is significantly increased only when multiple factors deviate together; when any input factor is abnormally missing, it should be replaced with a conservative value; when the vulnerability change exceeds 0.15 within two consecutive control cycles, a first-order filter can be added to avoid frequent oscillations in the target output of the fuel cell.

[0084] In practical implementation, the prediction time domain, update frequency, and degradation correction rules for the available power of the fuel cell should be set such that the prediction time domain is preferably set to 0.5 to 2 seconds, corresponding to the short-term future interval covered by the current rolling window; the available power of the fuel cell should be updated once in each control cycle, and the effective estimate of the previous control cycle should be used as the initial value for smoothing; if the average temperature of the stack is too high, the pressure fluctuation is abnormal, or the oxygen supply is insufficient, degradation reduction should be introduced based on the current normalized fuel cell output; when the system is in stable flight and the risk quantities are low, the available power can be close to the current normalized fuel cell output.

[0085] In practical implementation, to aggregate the predicted load demand, the supercapacitor's supported power, and future energy gap indicators, the predicted load demand needs to be calculated for each discrete point in the preview path segment based on its speed, curvature, vertical maneuvering, and trajectory tracking correction. The supercapacitor's supported power should be determined by comprehensively considering the current state of charge, terminal voltage, temperature, allowable discharge current, equivalent internal resistance, and protection threshold. The future energy gap indicator should preferentially take the maximum value of the gap ratio of each discrete point in the preview window, and use the moving average result of the most recent three periods as the final control input. When the difference between the maximum value and the average value is too large, the peak value and the average value can be recorded simultaneously for path reconstruction and power allocation.

[0086] Preferably, it generates the target output of the supercapacitor and the target output of the fuel cell based on the future energy supply gap index and the anti-gravity drainage exposure memory, and simultaneously adaptively generates hydrogen supply regulation commands and continuous hydrogen discharge duty cycles, including: The future energy supply gap indicator With respect to the normalized supercapacitor's available state of charge Multiply to obtain a first product, divide the first product by the first product, and subtract the normalized path segment power vulnerability. The difference and the numerical stability term The sum of these three factors yields the supercapacitor support ratio. : Support ratio of the supercapacitor The path power demand index at the current control time Multiplying the results yields the normalized target output of the supercapacitor, which is then used as the target output of the supercapacitor. Subtract the supercapacitor support ratio from one. The difference between the path power demand index at the current control moment and the path power demand index at the current control moment. Multiplying the results yields the normalized target output of the fuel cell, which is then used as the target output of the fuel cell. : The normalized fuel cell target output Multiply by one and the amount of memory exposed by the anti-gravity drainage The sum, divided by one, is the normalized pressure risk mentioned above. and the normalized temperature risk The sum of these three factors yields a normalized hydrogen supply regulation command, which serves as the hydrogen supply regulation command. : Expose the memory value of the anti-gravity drainage With the normalized stack current Multiply to obtain a third product, divide the third product by the third product, and subtract the normalized available state of charge of the supercapacitor. The difference and the numerical stability term The sum of these three factors yields the normalized hydrogen emission duty cycle as the continuous hydrogen emission duty cycle. : The first product is the result of multiplying the future energy supply gap index with the normalized available state of charge of the supercapacitor, which is used to characterize the basic weight of the current gap compensation undertaken by the supercapacitor.

[0087] The supercapacitor support ratio is an allocation ratio derived from a comprehensive consideration of future energy supply gap indicators, the available state of charge of supercapacitors, and the power vulnerability of the power transmission route.

[0088] The normalized supercapacitor target output is the dimensionless target power after the current path power demand index is allocated according to the supercapacitor support ratio.

[0089] The target output of a supercapacitor is the target output quantity that the supercapacitor channel ultimately needs to track.

[0090] The normalized target output of a fuel cell is the dimensionless target power allocated to the fuel cell after deducting the portion borne by the supercapacitor from the current path power demand index.

[0091] The target output of a fuel cell is the target output quantity that the fuel cell channel ultimately needs to track.

[0092] The normalized hydrogen supply regulation command is a dimensionless hydrogen supply control quantity obtained by comprehensively considering the fuel cell target output, the anti-gravity drainage exposure memory, the normalized pressure risk, and the normalized temperature risk.

[0093] The hydrogen supply regulation command is the actual control objective issued to the hydrogen supply components.

[0094] The third product is the product of the antigravity drainage exposure memory and the normalized stack current, used to characterize the intensity of current hydrogen emission demand.

[0095] The normalized hydrogen emission duty cycle is a dimensionless hydrogen emission duty cycle generated based on the third product, the normalized supercapacitor available state of charge, and the numerical stability term. It is used to describe the intensity of hydrogen emission execution.

[0096] The continuous hydrogen venting duty cycle is the actual duty cycle adjustment target used by the hydrogen venting component during continuous control, which is used to control the opening and closing ratio of the hydrogen venting valve over a period of time.

[0097] In practical implementation, regarding the boundary, smoothing, and failure protection of the supercapacitor support ratio, the supercapacitor support ratio needs to be limited to between 0 and 1 to avoid over-allocation or negative allocation. When the future energy supply gap index rises rapidly and the supercapacitor's available state of charge is sufficient, the support ratio can be quickly increased, but the single-cycle increment should not exceed 0.1. When the supercapacitor temperature is too high, the terminal voltage is too low, or the management unit reports current limiting, the support ratio should be forcibly reduced. When the fuel cell is stable and the power vulnerability of the path segment is low, the support ratio should be gradually reduced back to a lower level to retain subsequent transient buffering capacity. If the supercapacitor channel fails, the support ratio should be directly set to 0, and the path risk weight and deceleration level should be increased simultaneously.

[0098] In practical implementation, the conversion from normalized values ​​to actual output values ​​for supercapacitor and fuel cell target outputs requires first converting the normalized target value into an actual power target based on their respective rated power, rated current, and safe operating range. Then, the power target is converted into a current target or duty cycle target based on the current bus voltage. The supercapacitor target output should have a ramp rate limit to avoid bus impact. The fuel cell target output should have a gentler slope limit to prevent the hydrogen and oxygen supply systems from following too slowly. When the sum of the two exceeds the total propulsion demand, maintaining bus stability should be prioritized, and adjustments should be made according to the level of risk.

[0099] In practice, for the actuators and safety constraints corresponding to the hydrogen supply regulation command, the target opening degree of the proportional valve, the target pressure of the pressure reducing valve, or the target opening degree of the mass flow control valve corresponding to the hydrogen supply regulation command needs to be updated in conjunction with the current target output of the fuel cell stack, the fuel cell stack pressure risk, and the temperature risk. When the hydrogen supply pressure is close to the upper limit, the hydrogen supply regulation command should not be adjusted upward. When the hydrogen supply pressure is close to the lower limit and the fuel cell stack still needs to output higher power, the hydrogen supply target can be increased first, but the rate of change must be limited. If an increase in the risk of hydrogen leakage is detected, the hydrogen supply regulation command should immediately enter the conservative mode.

[0100] In practical implementation, for the valve control strategy and conduction constraints corresponding to the continuous hydrogen venting duty cycle, the normalized hydrogen venting duty cycle needs to be converted into the valve opening and closing ratio within a fixed control window, with the control window preferably between 0.5 seconds and 2 seconds. Each hydrogen venting action should meet the minimum conduction time and minimum shutdown time, preferably not less than 0.05 seconds and 0.2 seconds, respectively. When the water production intensity is high, the exposure memory of anti-gravity drainage is high, and the stack current is large, the hydrogen venting duty cycle should be increased. When the available state of charge of the supercapacitor is too low, excessively frequent hydrogen venting should be avoided to prevent additional power fluctuations. If the oxygen supply sufficiency is still not improved after two consecutive hydrogen ventings, load reduction or local path reconstruction should be triggered simultaneously.

[0101] Preferably, the exposure memory of the anti-gravity drainage and the future energy supply gap index are incorporated into the path evaluation cost, and the reconstructed local path and velocity distribution along the path are generated by optimizing the candidate path set, including: Obtain the candidate path set Candidate Path The single-point normalized obstacle distance corresponding to the single point Subtract the single-point normalized obstacle distance from one. The single-point normalized barrier approximation degree was calculated. Obtain the candidate path The single-point normalized local hydrogen concentration risk corresponding to the above single point Single-point normalized temperature risk Single-point normalization pressure risk Subtract the single-point normalized local hydrogen concentration risk from the above. The difference, minus the single-point normalized temperature risk The difference between the two values ​​is one minus the single-point normalized pressure risk. The differences are multiplied consecutively, and the single-point system security risk is obtained by subtracting the product from the result. : For each of the candidate paths Subtract the single-point normalization barrier approximation degree from one. The difference, minus the corresponding amount of anti-gravity drainage exposure memory. The difference, minus the corresponding future energy supply gap indicator The difference and a subtraction of the single-point system security risk The differences are multiplied consecutively, the fourth root is taken, and then the result is subtracted from the root to obtain the single-point cost of the candidate path. : In the candidate path set Selecting a small element along the normalized arc length The candidate path with the minimum single-point cost integral is selected as the reconstructed local path. : Obtain the distribution of the reconstructed local paths Reference speed along the path , along the process of anti-gravity drainage exposure memory Obstacle Approach Degree and indicators of future energy supply gaps along the route The reference speed along the path Multiply by one and subtract the amount of memory exposed by the anti-gravity drainage along the route. The difference and a subtraction of the approach degree along the path obstacle The difference, divided by a factor equal to the future energy supply gap index along the route. The sum of the values ​​is used to update the reconstructed local path. The speed distribution along the path mentioned above : The cost of path evaluation is a measure of the path's merits and demerits, taking into account factors such as approaching obstacles, exposure to counter-gravity drainage, future energy supply gaps, and system security risks.

[0102] The candidate path set is a collection of multiple flyable paths generated by the local planner at the current moment.

[0103] A candidate path is a single local path scheme in the set of candidate paths.

[0104] The single-point normalized obstacle distance is a dimensionless quantity that maps the distance between the UAV and the nearest obstacle at a single point on the candidate path, and is used to characterize the near-obstacle state at that point.

[0105] The single-point normalized obstacle proximity is a dimensionless degree of proximity obtained by converting the single-point normalized obstacle distance, and is used to characterize the near-obstacle risk at that point.

[0106] The single-point normalized local hydrogen concentration risk is a dimensionless quantity representing the local hydrogen anomaly risk at a single point on a candidate path, used to characterize the hydrogen safety risk when passing through that spatial location.

[0107] Single-point normalized temperature risk is a dimensionless risk quantity related to the thermal state of the fuel cell stack at a single point in the candidate path, used to characterize the thermal safety pressure under that path state.

[0108] Single-point normalized pressure risk is a dimensionless risk quantity related to the hydrogen or gas supply pressure state at a single point on a candidate path, used to characterize the safety risk on the pressure side.

[0109] The single-point system safety risk is a comprehensive safety risk quantity obtained by integrating the single-point normalized local hydrogen concentration risk, single-point normalized temperature risk, and single-point normalized pressure risk. It is used to measure the systemic safety level of a single point on a candidate path.

[0110] The single-point cost of the candidate path is the single-point evaluation value that integrates the single-point normalized obstacle approximation degree, the memory of anti-gravity drainage exposure, the future energy supply gap, and the single-point system safety risk.

[0111] The normalized arc length differential is the standardized path offset length used when integrating candidate paths. A preferred value range is 0.01 to 0.05 for the normalized arc length, as this range balances local risk resolution and real-time computational complexity.

[0112] Reconstructing a local path is the optimal local path obtained by selecting the best candidate path from the candidate path set by integrating the path evaluation cost.

[0113] The reference speed along the path is the baseline speed distribution given by the upper-level task or the original planner for reconstructing the local path.

[0114] The exposure memory of antigravity drainage along the path is the distribution of the exposure memory of antigravity drainage after unfolding along the reconstructed local path.

[0115] Obstacle proximity along the path is the distribution of the degree of proximity of obstacles to each point along the reconstructed local path.

[0116] The future energy supply gap index along the route is the predicted distribution of future energy supply gaps at each point along the local reconstruction path.

[0117] The path-based speed distribution is the actual speed planning result updated by combining the path-based reference speed, the path-based anti-gravity drainage exposure memory, the path-based obstacle approach degree, and the path-based future energy supply gap index. It is used to guide the UAV's forward speed control along the path.

[0118] In practice, regarding the generation and feasibility constraints of the candidate path set, multiple candidate paths need to be generated near the local reference path, starting from the current position and velocity direction of the UAV, using a combination of lateral offset, longitudinal altitude fine-tuning, and curvature sampling. The candidate path set should preferably include 5 to 21 paths to balance search capability and real-time performance. The lateral spacing between adjacent candidate paths can be set to 1 to 2 times the diagonal dimension of the UAV, and the altitude spacing can be set according to the safety vertical margin. Each candidate path must meet the constraints of minimum turning radius, maximum climb rate, maximum descent rate, and obstacle safety distance. If a candidate path touches a no-fly zone, communication loss zone, or known high hydrogen concentration danger zone, it should be directly eliminated.

[0119] In practical implementation, the evaluation of single-point system safety risks and single-point costs of candidate paths requires that the single-point normalized obstacle proximity be obtained by mapping the distance from the point to the nearest obstacle; the smaller the distance, the higher the proximity. The single-point system safety risk should integrate the single-point normalized local hydrogen concentration risk, single-point normalized temperature risk, and single-point normalized pressure risk into a unified risk quantity. The single-point cost of candidate paths should simultaneously consider the degree of proximity to obstacles, adverse drainage memory, future energy supply gap, and system safety risk, and limit the result to between 0 and 1. When any risk at a single point exceeds the emergency threshold, the cost of that point can be directly set to a high value.

[0120] In practice, for cost integration and optimization along normalized arc-length infinitesimals, each candidate path needs to be discretely sampled using a unified normalized arc-length infinitesimal, and the total path cost needs to be accumulated using rectangular or trapezoidal integration. Sampling can be appropriately densified near the takeoff point, in areas with dense obstacles, and in areas with expected high loads to improve local risk resolution. The smaller the total path integral value, the lower the overall risk and energy pressure. When the integral values ​​of two candidate paths are close, the path with gentler curvature and smaller speed changes should be selected first. If the optimal candidate path deviates too much from the current execution path, continuity constraints should be added to avoid unnecessary large turns.

[0121] In practice, for the indicators of exposure memory of anti-gravity drainage along the route, obstacle approach degree along the route, and future energy supply gap along the route, it is necessary to recursively extrapolate the exposure memory of anti-gravity drainage along the route point by point, combining the expected attitude changes of the candidate path, the expected stack current, and the planned hydrogen emission behavior; the obstacle approach degree along the route can be obtained point by point by querying the results of local map, laser point cloud grid occupancy, and body envelope expansion; the future energy supply gap along the route can be calculated based on the predicted load demand and corresponding available energy supply capacity of each discrete point; if there is a significant continuous pitch change or long-term hovering in a certain section of the path, the exposure memory of anti-gravity drainage along that section should be increased accordingly.

[0122] In practice, for the smoothing, speed limiting and dynamic flight constraints of the velocity distribution along the flight path, the updated velocity distribution along the flight path needs to meet the limits of maximum acceleration, maximum deceleration and maximum jerk to avoid sudden changes in speed commands; active speed limiting should be implemented in the near obstacle zone, the high value zone of the power supply gap and the high value zone of the anti-gravity drainage exposure; in the open area and when energy is sufficient, the velocity can be gradually restored to the reference velocity along the flight path; the rate of change of velocity between adjacent discrete points can be limited by first-order or second-order smoothing.

[0123] Preferably, updating the center of gravity position vector based on mass migration caused by fluid material consumption, and generating a feedforward compensation torque based on the center of gravity position vector to correct the rotor target command, includes: The normalized hydrogen storage mass obtained by relative normalization based on the initial state The equivalent mass of stagnant water is derived from the water production intensity and the continuous hydrogen evacuation duty cycle. Normalized supercapacitor quality Normalized load mass and normalized body baseline mass , respectively with the corresponding normalized local centroid position vector , , , and Multiply and sum as the mass torque; the normalized hydrogen storage mass The equivalent mass of the stagnant water The normalized supercapacitor mass The normalized load mass The normalized body reference mass and the numerical stability term The sum of the mass torques is the total system mass; the updated center of gravity position vector is obtained by dividing the sum of the mass torques by the total system mass. : The centroid position vector The resultant force vector required for normalized trajectory tracking Perform a cross product to obtain the feedforward compensation torque. : The normalized control allocation matrix is ​​updated based on the current achievable thrust of each rotor. The feedforward compensation torque With normalized trajectory tracking feedback torque After addition, the normalized target thrust Combine and left-multiply the pseudo-inverse of the normalized control allocation matrix. The normalized rotor target command is derived as the rotor target command. : Normalized hydrogen storage mass is the result of normalizing the remaining hydrogen mass of the current hydrogen storage system relative to the initial baseline state, and is used to characterize the impact of changes in fuel reserves on the center of gravity.

[0124] The initial state is the baseline configuration state after hydrogen refueling, payload assembly, drainage checks, and power preparation are completed before takeoff. Preferably, it is the static calibration state before the actual mission, and all mass, center of gravity, and energy parameters should be recursively referenced to this state.

[0125] Water production intensity is the amount of liquid water generated per unit time by a fuel cell under the current load.

[0126] The equivalent mass of retained liquid water is derived from the production intensity and continuous hydrogen evacuation occupancy process. It is used to characterize the additional impact of liquid water that is not discharged in time on the center of gravity and stack state.

[0127] The normalized supercapacitor mass is the result of normalizing the supercapacitor component mass relative to the system's baseline mass. It can be obtained through ground-based weighing calibration and subsequent conversion.

[0128] Normalized load mass is the result of normalizing the mission load mass relative to the system baseline mass. It can be obtained through actual weighing after the load is installed.

[0129] The normalized reference mass of the unit is the normalized mass of the fixed parts of the unit, excluding hydrogen storage, liquid water, and variable loads. It can be obtained by weighing the entire unit empty and then subtracting the variable mass.

[0130] The normalized local centroid position vector of hydrogen storage is the scale-normalized vector representing the local centroid position of the hydrogen storage component relative to a reference point on the main body. It can be obtained through 3D modeling, suspension measurement, or multi-point weighing inversion.

[0131] The normalized local centroid position vector of the liquid water is the normalized vector of the position of the equivalent mass of the retained liquid water relative to the reference point of the organism. It is preferably a position value near the geometric center of the fuel cell stack's water collection area and drainage channel, where the retained liquid water is mainly concentrated.

[0132] The normalized local centroid position vector of a supercapacitor is a vector obtained by normalizing the local centroid position of the supercapacitor module. It can be obtained through supercapacitor module installation measurement and 3D assembly calibration.

[0133] The normalized local centroid position vector of the load is the vector obtained by normalizing the local centroid position of the task load. It can be obtained through load installation measurement, suspension method, or three-dimensional model calibration.

[0134] The normalized local center of gravity vector of the aircraft is the vector obtained by normalizing the local center of gravity of the fixed parts of the aircraft. It can be obtained by testing the overall center of gravity of the aircraft after removing variable masses.

[0135] The sum of mass torques is the combined torque obtained by multiplying the mass of each component with the corresponding local center of gravity position vector.

[0136] The total system mass is the combined mass of the hydrogen storage mass, the equivalent mass of the stagnant liquid water, the supercapacitor mass, the payload mass, and the reference mass of the system body.

[0137] The normalized resultant force vector required for trajectory tracking is the vector of the total resultant force required by the UAV to meet the current trajectory tracking target after scale normalization.

[0138] The feedforward compensation torque is a compensation torque determined by the center of gravity position vector and the resultant force vector required for normalized trajectory tracking. It is used to correct attitude control deviations caused by center of gravity offset in advance.

[0139] The current achievable thrust of each rotor is the maximum usable thrust that each rotor can safely output under the current speed, voltage, air density, and protection limits, and is used to update the control allocation capability boundary.

[0140] The normalized control allocation matrix is ​​a dimensionless matrix that describes the mapping relationship between rotor thrust and overall target thrust and target torque, and is used to complete multi-rotor control allocation.

[0141] The normalized trajectory tracking feedback torque is the result of the feedback torque generated by the closed-loop trajectory tracking controller based on attitude error and angular velocity error, after scale normalization, and is used to correct real-time deviations.

[0142] Normalized target thrust is a normalized expression of the total thrust required to satisfy the current trajectory and altitude control.

[0143] The pseudo-inverse of the normalized control allocation matrix is ​​the generalized inverse of the current control allocation matrix, used to solve for rotor target commands under multi-actuator redundancy conditions.

[0144] The normalized rotor target command is a dimensionless control quantity of each rotor obtained by control allocation based on the target thrust and target torque.

[0145] The rotor target command is the actual execution command issued to the motor controller or rotor speed controller to drive the multi-rotor assembly to output the desired thrust.

[0146] In practical implementation, the recursive model for normalized hydrogen storage mass, water production intensity, and equivalent mass of stagnant liquid water needs to be established. This requires calculating the hydrogen consumption based on the actual output power of the fuel cell, the lower heating value of hydrogen, and the system efficiency, and then decreasing it periodically. The water production intensity should be calculated based on the stack current and electrochemical reaction relationship, and corrected in conjunction with the stack operating temperature and load status. The equivalent mass of stagnant liquid water should be recursively derived from the stagnant amount in the previous cycle, the current water production, and the hydrogen removal amount in the current cycle. When the hydrogen removal duty cycle increases, the growth rate of the equivalent mass of stagnant liquid water should slow down or decrease.

[0147] In practice, the calibration of the mass of each component and the normalized local center of gravity position vector requires weighing and three-dimensional position measurement of the hydrogen storage system, supercapacitor assembly, mission payload, and fixed parts of the machine body after the entire assembly is completed. The local center of gravity position can be obtained through three-dimensional modeling, suspension measurement, or multi-support weighing inversion. All mass parameters are uniformly normalized according to the system reference mass, and all position parameters are uniformly normalized according to the reference length of the machine body. When the load is changed or the installation position of the hydrogen storage cylinder is changed, the corresponding local center of gravity position vector and mass parameters should be recalibrated.

[0148] In practical implementation, regarding the coordinate system consistency and amplitude limiting coupling rules for the feedforward compensation torque, it is necessary to calculate the center of gravity position vector and the resultant force vector required for normalized trajectory tracking in the same machine system to avoid errors in torque direction caused by the mixing of coordinate systems. The obtained feedforward compensation torque should be limited separately along the three axes of roll, pitch, and yaw, with the upper limit of the limit preferably not exceeding 70% to 80% of the currently available control torque capability. When the feedforward compensation torque and the feedback torque are in opposite directions, they can be directly superimposed and canceled out. When the two are in the same direction and the combined result exceeds the execution limit, priority should be given to ensuring the roll and pitch components related to attitude stability.

[0149] In practice, the update rules for the current achievable thrust of each rotor and the normalized control allocation matrix need to be set in real time, taking into account motor speed, bus voltage, propeller aerodynamic characteristics, ambient air density, and ESC current limiting information. When a rotor's temperature is too high, the ESC current is limited, or the speed tracking is abnormal, the current achievable thrust of that rotor should be reduced. The normalized control allocation matrix should be dynamically updated based on the rotor layout, rotor lever arm, and the upper limit of the current achievable thrust of each rotor. For rotors with lower achievable thrust, a smaller weight can be introduced into the matrix to reduce their allocation burden.

[0150] In practical implementation, for pseudo-inverse solving, rotor saturation processing, and fault degradation strategy settings, the weighted least squares method should be used first to obtain the pseudo-inverse matrix to take into account both total thrust tracking and attitude moment tracking. When the solved rotor target command exceeds its allowable range, saturation clipping should be performed and the remaining control quantities should be reallocated. When a single rotor fault or severe performance degradation is detected, the rotor weight should be reduced from the control allocation or it should be directly removed, and the flight mode should be entered simultaneously. In the degradation mode, altitude and roll-pitch stability should be maintained first, and yaw performance can be appropriately compromised.

[0151] Preferably, the system's overall risk is assessed and an emergency takeover coefficient is generated. This emergency takeover coefficient is then used to smoothly fuse the reconstructed local path and the emergency landing path to generate the final command path, including: For candidate landing areas, the corresponding normalized flatness risk will be subtracted. The difference, minus the corresponding normalized net short risk The difference and the corresponding normalized wind direction risk minus one Multiply the differences and subtract the product to obtain the landing non-feasibility. : Subtract the normalized barrier distance from 1 Obtain the current normalization barrier approximation degree ; Subtract the normalized local hydrogen concentration risk from 1 The difference, minus the normalized temperature risk The difference, minus the normalized pressure risk The difference, minus the future energy supply gap indicator The difference, minus the current normalization barrier approximation. The difference and a subtraction of the landing infeasibility The differences are multiplied consecutively, and the result of the product is taken as the sixth root. The comprehensive risk of the system is then obtained by subtracting the result from one. : The system's overall risk Divide by its normalized remaining executable energy and the numerical stability term The emergency takeover coefficient is obtained by summing the results. : Including terrain cost The optimal emergency landing path is selected under the evaluation system. : Subtract the emergency takeover coefficient from 1 The difference multiplied by the reconstructed local path In addition to the aforementioned emergency takeover coefficient With the emergency landing path The product of these terms yields the final command path of continuous fusion. : The overall system risk is a total risk indicator formed by the integration of hydrogen safety, thermal safety, pressure safety, energy supply gap, near-obstacle risk, and landing infeasibility.

[0152] The emergency takeover coefficient is a fusion weight calculated based on the overall system risk and remaining executable energy. It is used to determine the degree to which current control is transferred from the reconstructed local path to the emergency landing path.

[0153] Candidate landing areas are a set of ground or platform areas in the current environment that are identified as potentially suitable for an emergency landing, and are used to generate target areas for emergency landing paths.

[0154] Normalized smoothness risk is a dimensionless risk quantity derived from the surface undulation of the candidate landing area, used to characterize the landing stability risk.

[0155] Normalized headroom risk is a dimensionless risk quantity formed by combining obstacles around the candidate landing area, overhead obstructions, and available space in the glide path. It is used to characterize the safety of the landing path.

[0156] Normalized wind direction risk is a dimensionless risk quantity derived from the degree of matching between the local wind direction in the candidate landing area and the approach direction of the UAV. It is used to characterize the aerodynamic disturbance risk during landing.

[0157] Landing infeasibility is a measure of landing execution difficulty obtained by integrating flatness risk, airspace risk, and wind direction risk. It is used to characterize whether a candidate landing area is suitable for immediate landing.

[0158] The current normalized obstacle proximity is a dimensionless quantity representing the degree of proximity between the UAV and nearby obstacles at the current moment, used to reflect the collision pressure under the current flight conditions.

[0159] Normalized remaining executable energy is the normalized result of the remaining energy of the UAV that can still be used for flight, obstacle avoidance and landing while maintaining safety redundancy. It is used to constrain the intensity of emergency takeover.

[0160] The emergency landing path candidate set is a set of multiple feasible emergency paths leading to each candidate landing area, used for path optimization in emergency mode.

[0161] Terrain cost is a path evaluation metric composed of ground slope, roughness, obstacle distribution, accessibility, and surface type, used to screen for paths more suitable for emergency descent.

[0162] An emergency landing path is an emergency flight path selected by combining the risks of candidate landing areas and the terrain cost. It is used to guide UAVs to land safely in high-risk situations.

[0163] The final command path is the final execution path obtained by continuously fusing the reconstructed local path and the emergency landing path according to the emergency takeover coefficient.

[0164] In practice, the evaluation methods for candidate landing areas, as well as for risks related to flatness, clearance, and wind direction, should be set up as follows: candidate landing areas should be jointly identified by downward-looking laser ranging, depth vision, 3D topographic maps, and historical high-precision maps; flatness risk should be obtained by comprehensively mapping the elevation undulations, local slopes, and surface roughness within the area; clearance risk should be determined by comprehensively considering the height of obstacles above the landing area, the distance between surrounding obstacles, and the width of the approach corridor; and wind direction risk should be determined by comprehensively considering the angle between the local wind direction and the UAV's planned approach direction, as well as the intensity of gusts.

[0165] In practice, the thresholds, weights, and limit rules for landing infeasibility and system comprehensive risk should be set. Landing infeasibility should be used to characterize whether a candidate landing area has the basic conditions for immediate landing, and system comprehensive risk should be used to characterize whether it is still safe to continue executing the normal path. Both should be limited to between 0 and 1. When the system comprehensive risk is below 0.3, the execution of the normal local path can be maintained. When the system comprehensive risk is above 0.6, the emergency takeover coefficient should be gradually increased. When the system comprehensive risk is above 0.8 and the remaining executable energy is insufficient, the nearest low-risk landing area should be selected first.

[0166] In practical implementation, the definition and update rules for normalized remaining executable energy should be determined by the remaining releasable hydrogen energy, the available energy of the supercapacitor, and the reserved safety redundancy. Among them, the remaining releasable hydrogen energy should deduct fuel cell efficiency losses and unusable tail energy, and the available energy of the supercapacitor should deduct the reserved portion below the minimum allowable state of charge. This value should be updated according to the current power consumption in each control cycle and normalized according to the maximum available mission energy of the system. When environmental deterioration leads to a significant increase in flight power, the rate of decrease of remaining executable energy should be accelerated simultaneously.

[0167] In practice, the establishment of an emergency landing path candidate set and an evaluation system that includes terrain costs requires generating multiple emergency landing path candidates between the current UAV position and the target point, with each candidate landing area as the target point. During generation, constraints such as minimum turning radius, minimum glide safety altitude, obstacle clearance, and maximum descent rate must be met. Terrain costs should comprehensively consider ground slope, vegetation cover, water surface, gravel ground, building density, and degree of undulation. The total evaluation value of the emergency landing path should simultaneously consider obstacle approach, local hydrogen concentration risk, and terrain costs.

[0168] In practical implementation, for the path smooth fusion and switching protection mechanism after emergency takeover, the final command path needs to be obtained by continuously weighting and fusing the reconstructed local path and the emergency landing path according to the emergency takeover coefficient. During the fusion process, the continuity of position, heading, and curvature must be maintained. When the emergency takeover coefficient continues to increase, the path deflection within a unit period should be limited to prevent the aircraft from swaying abruptly. When the emergency takeover coefficient exceeds 0.7, the newly added optimization results of the normal path can be frozen, and only the emergency landing path should be retained as the dominant path. If the overall system risk decreases in a short period of time, but has not yet stabilized for several consecutive control cycles, the emergency fusion state should not be exited immediately to avoid back-and-forth switching.

[0169] It should be noted that the input and output parameters in the calculation formulas of this application are all dimensionless calculations performed through normalization processing. The formulas are all derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0170] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A flight control method specifically for hydrogen-powered drones, applied to hydrogen-powered drones comprising a proton exchange membrane fuel cell, a supercapacitor, an oxygen supply component, a hydrogen supply component, a hydrogen exhaust component, a sensor unit, and a multi-rotor component, characterized in that, include: The sensor state quantities of the sensor unit are acquired and synchronized, and dimensionless processing is performed according to the feasible value range of the rolling window to obtain a dimensionless state vector. The path power demand index is then calculated by feedforward. Based on the aircraft attitude of the hydrogen-powered UAV and the favorable drainage direction of the proton exchange membrane fuel cell, the anti-gravity drainage exposure memory quantity, which reflects the accumulation of unfavorable drainage, is extracted. The power vulnerability of future path segments is predicted by using the memory amount of the anti-gravity drainage exposure and the oxygen sufficiency in the sensed state quantity, and then the future energy supply gap index is output. Based on the future energy supply gap index and the memory capacity exposed by the anti-gravity drainage, the target output of the supercapacitor and the target output of the fuel cell are generated, and the hydrogen supply regulation command and the continuous hydrogen discharge duty cycle are generated simultaneously and adaptively. The exposure memory of the anti-gravity drainage and the future energy supply gap index are incorporated into the path evaluation cost, and the reconstructed local path and the speed distribution along the path are generated by optimizing the candidate path set. The center of gravity position vector is updated based on the mass migration caused by fluid material consumption, and a feedforward compensation torque is generated based on the center of gravity position vector to correct the rotor target command. Assess the overall risk of the system and generate an emergency takeover coefficient. Use the emergency takeover coefficient to smoothly merge the reconstructed local path and the emergency landing path to generate the final command path.

2. The flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 1, characterized in that, Acquire and synchronize the sensing state quantities of the sensor units, perform dimensionless processing according to the feasible value range of the rolling window to obtain a dimensionless state vector, and feedforward to calculate the path power demand index, including: Divide the current perceptible distance by the product of the current flight speed and the control cycle, add the sum of the numerical stability terms, and round up to determine the preview distance step count to generate the scrolling window; The asynchronously acquired sensing state quantities are aligned to the current control moment using time interpolation combined with the numerical stability term; The aligned sensing state variables are normalized according to the feasible value range of the scrolling window to form the dimensionless state vector. The dimensionless state vector includes normalized velocity, normalized tangential acceleration, normalized vertical acceleration, normalized trajectory tracking error, normalized path curvature, normalized roll angle, normalized pitch angle, normalized hydrogen pressure, normalized pressure risk, normalized stack average temperature, normalized temperature dispersion, normalized temperature risk, normalized stack current, normalized oxygen sufficiency, normalized supercapacitor available state of charge, normalized local hydrogen concentration risk, normalized obstacle distance, and normalized center of gravity position vector. The demand numerator is obtained by adding the product of the normalized velocity and the normalized tangential acceleration, the product of the square of the normalized velocity and the normalized path curvature, the normalized vertical acceleration, and the normalized trajectory tracking error. The sum of the demand numerator and the demand denominator is used as the demand denominator. The path power demand index is obtained by dividing the demand numerator by the demand denominator.

3. The flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 2, characterized in that, Based on the aircraft attitude of the hydrogen-powered UAV and the favorable drainage direction of the proton exchange membrane fuel cell, the anti-gravity drainage exposure memory, reflecting the accumulation of unfavorable drainage, is extracted, including: Subtract the product of the unit vector in the direction of gravity, the direction cosine matrix from the machine system to the fuel cell flow channel coordinate system, and the unit vector in the favorable drainage direction, and then divide the resulting difference by two to obtain the normalized anti-gravity drainage exposure. The negative control period is used as the independent variable of the natural exponential function to obtain the exponential value, thus yielding the memory decay coefficient. The normalized hydrogen emission process is obtained by dividing the normalized time since the last hydrogen emission by the sum of the normalized recovery time and the numerical stability term. The first product term is obtained by multiplying the antigravity drainage exposure memory of the previous cycle by the memory decay coefficient; the second product term is obtained by multiplying the difference between the memory decay coefficient, the normalized antigravity drainage exposure, the normalized stack current, the normalized hydrogen emission process, and the difference between the normalized oxygen supply adequacy; the third product term is obtained by adding the first product term and the second product term.

4. The flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 3, characterized in that, The power vulnerability of future path segments is predicted by utilizing the anti-gravity drainage exposure memory and the oxygen sufficiency in the sensed state quantities, thereby outputting future energy supply gap indicators, including: The temperature at each sensing point is corrected by sensitivity weighting using the local flow velocity estimate, and the difference between the corrected maximum and minimum values ​​is used to obtain the normalized temperature dispersion. The difference between the anti-gravity drainage exposure memory value and the difference between the normalized temperature dispersion value, the normalized hydrogen pressure and the normalized oxygen supply sufficiency value are continuously multiplied together. The fourth root of the product is then taken and subtracted from the fourth root value to obtain the normalized path segment power vulnerability as the power vulnerability. Multiply the difference between the normalized path segment power vulnerability and the current normalized fuel cell output to obtain the fuel cell available power in the predicted time domain. For each discrete point of the preview path segment, the corresponding path power demand index is calculated as the predicted load demand; the predicted load demand is divided by the available power of the fuel cell, the power supported by the supercapacitor, and the sum of the numerical stability terms to obtain the future energy supply gap index.

5. The flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 4, characterized in that, Based on the future energy supply gap indicator and the anti-gravity drainage exposure memory, the target output of the supercapacitor and the target output of the fuel cell are generated, and hydrogen supply regulation commands and continuous hydrogen discharge duty cycles are generated simultaneously and adaptively, including: The first product is obtained by multiplying the future energy supply gap index by the normalized available state of charge of the supercapacitor. The first product is then divided by the sum of the first product, the difference between the normalized path segment power vulnerability and the numerical stability term to obtain the supercapacitor support ratio. Multiplying the supercapacitor support ratio by the path power demand index at the current control moment yields the normalized supercapacitor target output, which is used as the supercapacitor target output; multiplying the difference between one and the supercapacitor support ratio by the path power demand index at the current control moment yields the normalized fuel cell target output, which is used as the fuel cell target output. Multiply the normalized fuel cell target output by one and the sum of the anti-gravity drainage exposure memory, then divide by one, the sum of the normalized pressure risk and the normalized temperature risk, to obtain the normalized hydrogen supply regulation command as the hydrogen supply regulation command. Multiply the anti-gravity drainage exposure memory amount by the normalized stack current to obtain a third product. Divide the third product by the sum of the third product, the difference in the normalized supercapacitor available state of charge, and the numerical stability term to obtain the normalized hydrogen emission duty cycle as the continuous hydrogen emission duty cycle.

6. The flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 5, characterized in that, The exposure memory of the anti-gravity drainage and the future energy supply gap index are incorporated into the path evaluation cost. From the candidate path set, the reconstructed local path and the velocity distribution along the path are generated by optimization, including: Obtain the single-point normalized obstacle distance corresponding to a single point on a candidate path in the candidate path set, and calculate the single-point normalized obstacle approximation degree by subtracting the single-point normalized obstacle distance from the single point. Obtain the single-point normalized local hydrogen concentration risk, single-point normalized temperature risk, and single-point normalized pressure risk corresponding to a single point on the candidate path, and continuously multiply the difference between the single-point normalized local hydrogen concentration risk, the single-point normalized temperature risk, and the single-point normalized pressure risk by subtracting the difference from the single point normalized local hydrogen concentration risk, the single-point normalized temperature risk, and the single-point normalized pressure risk by subtracting the result of the product from the single point to obtain the single-point system safety risk. For each candidate path, the difference between the single-point normalized obstacle approximation degree, the difference between the corresponding anti-gravity drainage exposure memory, the difference between the corresponding future energy supply gap index, and the difference between the single-point system safety risk are multiplied consecutively, and the fourth root is taken. Then, the result is subtracted from the fourth root to obtain the single-point cost of the candidate path. The candidate path with the smallest single-point cost integral along the normalized arc length infinitesimal element is selected from the candidate path set as the reconstructed local path. The following parameters are obtained: the along-path reference speed, the along-path anti-gravity drainage exposure memory, the along-path obstacle approach, and the along-path future energy supply gap index. The along-path reference speed is multiplied by the difference between the along-path anti-gravity drainage exposure memory and the difference between the along-path obstacle approach, and then divided by the sum of the along-path future energy supply gap index to obtain the along-path speed distribution updated on the reconstructed local path.

7. A flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 6, characterized in that, The center of gravity position vector is updated based on the mass migration caused by fluid material consumption, and a feedforward compensation torque is generated based on the center of gravity position vector to correct the rotor target command, including: The normalized hydrogen storage mass obtained by relative normalization based on the initial state, the equivalent mass of the stagnant liquid water obtained by recursion based on the water production intensity and the continuous hydrogen discharge duty cycle, the normalized supercapacitor mass, the normalized load mass, and the normalized body reference mass are multiplied by the corresponding normalized local centroid position vector and summed to obtain the mass torque sum. The normalized hydrogen storage mass, the equivalent mass of the stagnant liquid water, the normalized supercapacitor mass, the normalized load mass, the normalized body reference mass, and the numerical stability term are added to obtain the total system mass. The mass torque sum is divided by the total system mass to obtain the updated centroid position vector. The feedforward compensation torque is obtained by cross-productting the center of gravity position vector with the resultant force vector required for normalized trajectory tracking. Based on the updated normalized control allocation matrix of the current achievable thrust of each rotor, the feedforward compensation torque and the normalized trajectory tracking feedback torque are added together and combined with the normalized target thrust. Then, the normalized rotor target command is obtained by left-multiplying the pseudo-inverse matrix of the normalized control allocation matrix.

8. A flight control method for a hydrogen-powered unmanned aerial vehicle according to claim 7, characterized in that, Assess the overall system risk and generate an emergency takeover coefficient. Use this emergency takeover coefficient to smoothly merge the reconstructed local path and the emergency landing path to generate the final command path, including: For a candidate landing area, the difference between 1 and the corresponding normalized smoothness risk, the difference between 1 and the corresponding normalized net air risk, and the difference between 1 and the corresponding normalized wind direction risk are multiplied together, and the landing infeasibility is obtained by subtracting the product from 1. Subtracting the normalized obstacle distance from 1 yields the current normalized obstacle proximity; multiplying the difference between 1 and ... The emergency takeover coefficient is obtained by dividing the overall system risk by the sum of the normalized remaining executable energy and the numerical stability term. The optimal emergency landing path is selected based on an evaluation system that includes terrain costs; Multiply the difference between 1 and the emergency takeover coefficient by the reconstructed local path, and add the product of the emergency takeover coefficient and the emergency landing path to obtain the continuously fused final command path.

Images